A graph-based workflow for extracting grain-scale toughness from meso-scale experiments

Abstract

We introduce a novel computational framework designed to allow an experimentalist to extract the local toughness associated with crack growth along an individual grain boundary from fracture experiments on standard specimens. The proposed framework relies on the ability of a graph neural network to perform high accuracy predictions of the micro-scale material toughness, utilizing a limited size dataset that can be obtained from standard fracture experiments. We analyze the functionality of the different components of the framework, focusing on the minimum size of the dataset required for the network training following an initial training cycle on a synthetic dataset. We demonstrate the high efficiency of the algorithm in accurately predicting the local (>68%) and global (>95%) crack growth toughness in scenarios where the toughness dependence on the microstructural parameters is substantially different than the one used for the synthetic dataset. The merit of the proposed framework arises from its flexibility: it is not limited to a specific mathematical description of the toughness at the micro-scale and simplicity: the required experiments are rather standard. As such, the proposed framework enables the study of a wide range of material microstructure properties and the investigation of their influence on the material toughness.